Classifications

• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
  • H04L1/00—Arrangements for detecting or preventing errors in the information received
  • H04L1/12—Arrangements for detecting or preventing errors in the information received by using return channel
  • H04L1/16—Arrangements for detecting or preventing errors in the information received by using return channel in which the return channel carries supervisory signals, e.g. repetition request signals
  • H04L1/18—Automatic repetition systems, e.g. Van Duuren systems
  • H04L1/1829—Arrangements specially adapted for the receiver end
  • H04L1/1848—Time-out mechanisms
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
  • H04L12/00—Data switching networks
  • H04L12/02—Details
  • H04L12/16—Arrangements for providing special services to substations
  • H04L12/18—Arrangements for providing special services to substations for broadcast or conference, e.g. multicast
  • H04L12/1863—Arrangements for providing special services to substations for broadcast or conference, e.g. multicast comprising mechanisms for improved reliability, e.g. status reports
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
  • H04L1/00—Arrangements for detecting or preventing errors in the information received
  • H04L2001/0092—Error control systems characterised by the topology of the transmission link
  • H04L2001/0093—Point-to-multipoint

Definitions

• This invention relates to a timer-based approach for supressing feedback during data transmission, and in particular but not exclusively to reliable multicast communication.
• Multicast communication can be used for the transmission of data in both a one-to-many situation (for example multimedia applications, tickertape feeds or bulk file transfer) or the many-to-many communication of data (for example conferencing or network gaming), and is increasingly gaining in importance with the deployment of multicast in the internet and with the increasing number of satellites.
• a sender 1 communicates data over a network 3 to a plurality of receivers 2.
• the network 3 could be any suitable network, such as for example, the internet, a local area network (LAN), satellite network, etc.
• the number of receivers 2 will depend on the particular multicast application, from just a few to possibly even millions of receivers, such as for cable-TV delivery, satellite or other wireless communication services.
• the number of receivers 2 also varies dynamically during a multicast transmission as receivers leave or join the multicast group.
• feedback messages In reliable multicast, used to guarantee the delivery of data to a group of receivers, feedback messages (FBMs) are returned from the receivers to either acknowledge correct receipt of data (positive acknowledgement messages known as ACKs), or loss of data (negative acknowledgement messages known as NACKs) .
• FBMs feedback messages
• ACKs positive acknowledgement messages
• NACKs loss of data
• a timer distribution function suitable for use with one particular multicast application may be entirely unsuitable for use with another multicast application, due to the varying scalability and latency requirements.
• multimedia applications can require both scalability and low latency.
• Low latency is also a requirement for collaborative applications such as data-conferences (whiteboarding), although the scaling requirements are more modest (less than 100 participants) .
• Collaborative applications like this will require, for example, latency of less than 400msec so that responses do not cause discomfort to the human participants.
• Message streaming applications such as tickertape and news feeds often require both low latency and scalability to thousands (or possibly millions) of receivers.
• Tickertape feeds to brokerage houses need to be particularly timely because the information loses value greatly as time passes, and there is also the need for strict reliability.
• bulk data delivery may have no specific latency requirement, and can be scheduled for delivery during the night when the traffic on a network is reduced. Strict reliability is usually the main concern for this application, with the need to ensure that a complete set of data is transferred correctly. However, even in bulk data delivery applications, it can sometimes be necessary to receive the data almost immediately, and therefore the latency requirements can vary widely.
• a method for selecting a value of one or more parameters of a timer function for use by a receiver for delaying feedback in a multicast system comprising: finding the one or more parameter values which minimise an expression defined as a function of the parameters, the expression comprising at least two terms, where one term relates to the expected number of feedback messages generated by receivers in the multicast system and the second term relates to the expected extra latency of the feedback due to the timer function.
• the receiver can then dealy sending a feedback message by a time period determined in relation to the timer function, and in the event that the receiver detects during the time period that a substantially duplicate feedback message was sent by another receiver then cancel sending its own feedback message.
• a method for selecting a timer function for use by a receiver for delaying feedback in a multicast system comprising the steps of : for each of at least two timer functions, minimising with respect to one or more parameters of the timer function an expression comprising at least two terms, where one term relates to the expected number of feedback messages generated by receivers in the multicast system and the second term relates to the expected extra latency of the feedback due to the timer function; and comparing the values of the minimized expressions for the " timer functions.
• Embodiments of the invention advantageously allow an optimum timer function to be chosen for use in a multicast feedback mechanism, taking into account both the feedback and latency requirements for the multicast.
• parameters are selected to optimize an already-decided form of timer function.
• the method is used to simultaneously optimize and select one of a number of different forms of timer function.
• the expression may also include means for weighting the relative importance of the first and second terms, i.e. the NACK suppression requirements versus the latency constraints.
• the relative weighting of these two terms can therefore be modified for different multicast applications to reflect different situations in which, for example, the latency constraint may be more or less important, and to calculate an optimum timer function accordingly.
• the second term (relating to the extra latency due to the timer-based feedback mechanism) may take the form of a function which has a maximum gradient corresponding to where the extra latency reaches a predefined maximum acceptable limit.
• a minimised solution to the expression is therefore forced to lie below this point, namely where the latency is less than the maximum acceptable value. This advantageously allows a solution to be found in which the extra latency does not exceed a predefined limit, which can be set for example by a user or stored within the multicast system.
• the embodiments may also include the step of recalculating the optimised parameters (and/or reselecting a timer function) in response to changes in multicast conditions.
• a timer function optimised for specific multicast conditions e.g. for a particular multicast group size, etc
• Embodiments may use as the timer distribution function, a shifted power law timer of the form
• Figure 1 is a schematic drawing showing multicast transmission
• Figure 2 is a flowchart showing the steps for optimising a parameterised timer function for use in NACK suppression for reliable multicast, in accordance with embodiments of the invention
• Figure 3 is a flowchart showing the steps peformed by a multicast system, according to a first embodiment of the invention
• Figure 4(a) and (b) are graphs showing the expected performance of two optimised timer functions in reliable multicast
• Figure 5 is a graph showing the expected performance in reliable multicast of a shifted power law timer function optimised under different latency constraints
• Figure 6 is a graph showing the expectged relative performance in reliable multicast of two optimised timer functions, showing the expected excess latency versus number of NACKs
• Figure 7 is a flowchart showing the steps peformed by a multicast system, according to a second embodiment of the invention.
• Parameters ⁇ a ⁇ for the timer function, and also possibly an indication of which timer function the receivers are to use, are sent to the receivers in multicast messages transmitted from the sender.
• a mathematical model of the timer-based feedback mechanism is developed below, in which the delay times between sender and the receivers and among receivers themselves are deterministic and are homogeneously distributed.
• Two performance measures which are considered are the expected number of NACK feedback messages E[X] and the excess latency due to the feedback mechanism E[M] .
• the excess latency is the time delay which is additional to the usual network delay time c for a data packet to travel from a receiver to the sender, and corresponds to the expected time for the expiry of the first timer.
• the number of potential NACK senders (for the same loss probability q ) is R « (l - ( ⁇ - q) R )R , which quickly approaches R for even very modest loss probabilities.
• the expected number of NACK senders in a real system should be between these two extreme cases, and can be evaluated exactly using the probability distribution for the number of losses.,, per multicast round. However, in this case we consider a worst-case scenario in which all receivers are potential NACK senders.
• E[X] and E[M] can be written as functionals of the timer probability distribution function f' (t) and the corresponding cumulative timer probability distribution function E' (t) where
• the excess latency due to the feedback mechanism is the expected time before the first timer expires.
• the expected excess latency can be expressed as
• E[ ] ⁇ dt ⁇ l- ⁇ (i)) R where we have made use of the property E(t-c) ⁇ l for t ⁇ c.
• E[X] is just the average number of receivers whose NACKs cannot be cancelled by the suppression mechanism since they were sent within the network delay time c. It provides a lower bound to minimum number of NACKS that can be achieved with a given timer distribution function.
• ⁇ E[X]+ w ⁇ (E[M ]) where w is a weight and Q(E[M]) ⁇ s a suitably chosen function relating to the excess latency.
• Minimization of the objective function ⁇ therefore optimizes the parameterised timer function in relation to simultaneously minimizing both the number of expected NACK feedback messages and the excess latency due to the feedback mechanism, and is used within the embodiments of the invention as described in relation to Figs.2-7.
• a first embodiment according to the invention comprises a multicast system which operates using minimization of the objective function ⁇ described above.
• a multicast sender initially collates inputs 20 electronically which are to be used during processing stage 30.
• a second input 22 is an estimate of R, the number of receivers in the multicast group.
• Input 23 is an estimate of the GRTT (greatest network round trip time), which may also be either a predicted value or the currently maintained value for GRTT during multicast.
• Input 24 comprises any other relevant constraints, such as latency or feedback requirements specific to the multicast application, for example, the maximum acceptable excess latency E°[M] ,
• the first processing step 25 is the setting of initial conditions (based on the inputs) .
• step 25 will include setting the weighting term w to an appropriate value such as 31n(i?) (as discussed later), and setting the timer period T based on GRTT (which for the approximation of homogeneous network delay times, is equivalent to 2c, c being the one-way network delay time).
• T should be set to be larger than c, to ensure that not all the receivers backoff timers expire before the first NACK is received by the sender or other receivers.
• Optimisation commences at step 26 by computing values for E[X] and E[M] , and at step 27 the objective function ⁇ is computed.
• the parameters of the timer function are adjusted iteratively and steps 26-28 repeated until the optimum solution has been reached (i.e. the objective function ⁇ has been minimised) .
• the optimised solution is output at step 31 .
• the output for a single input timer function will comprise the optimized parameter settings ⁇ a) opt to be used in that timer function.
• the optimum solution will be selected by direct comparison of the minimized values of ⁇ for each timer, and an indication of the selected timer function will be output together with the relevant optimised parameters for that timer function.
• the optimised output at step 31 then forms part of the multicast messages transmitted by the sender during multicast at step 1 0 (see Fig. 3).
• the sender periodically monitors the multicast conditions (step 40). For example, the sender maintains updated information relating to the group dynamics such as the size of the group (number of receivers, R) or the greatest round trip time GRTT.
• a test is applied to determine whether specific multicast condtions have changed (for example, whether the group size R has increased by a predefined amount) . If not, processing returns to step 40 until the next periodic check is performed during multicasting. However, if multicast conditions have changed then re-optimisation of the timer function will occur as follows.
• Updated inputs are collated at stage 20' and fed electronically into processing stage 30' .
• Processing stage 30' then repeats the optimisation of the timer function(s) in the same manner as described earlier with reference to stage 30 (Fig. 2).
• a revised optimum solution is calculated and output for use in the next multicast round (step 42) .
• the optimisation method incorporates both NACK minimisation requirements and latency constraints, both taken into account in the objective function ⁇ , and can be adapted to reflect specific multicast requirements by changing the input constraints (step 24).
• the inputs at step 24 will affect the initial conditions set at step 25 and can therefore change the form of the objective function ⁇ as appropriate.
• the input constraints might indicate that minimum NACK feedback is required under a condition of maximum tolerable excess latency E°[M] ,
• the objective function will include the second term ⁇ (E[ ]) relating to excess latency of
• ⁇ (E[M]) 1 l - l + exp( (E[ ] - E°[M])) as described earlier.
• a suitable value for w would be Cln(R) , (where C is a constant around 3) which maintains the two terms in ⁇ roughly of the same order since the expected feedback grows with ln(R) .
• Minimization of the objective function will find a solution under which the expected excess latency should not exceed the threshold E°[M] .
• a timer function optimized in this way is therefore ideally suited for use in a multicast situation where it is important to maintain a particular latency requirement.
• the weighting term could be set at w > 0 , and ⁇ (E[M]) replaced simply by E[M] .
• T is the timer period.
• Fig. 4(b) shows the corresponding values for the excess latency (in units of network delay time c).
• Fig. 4 shows that the shifted power-law timer outperforms the exponential timer here, resulting in both a lower number of expected NACKs and also a lower excess latency.
• a system of up to R - 10 6 receivers was considered, with the timer period fixed at T- 10c. From the graph it is clear that the optimised timer was able to achieve the required latency. In addition, the timer was able to achieve very efficient NACK suppression, with the corresponding number of NACKs always remaining below 8 (not shown).
• T cannot be freely adjusted as it must be chosen to be at least as large as the network delay time c, and preferably should be larger than the sender-receiver greatest round trip time (GRTT) .
• GRTT sender-receiver greatest round trip time
• FIG. 6 shows the comparative expected performance in reliable multicast of the shifted power law (SPL) and exponential (EXP) timer functions optimised using a range of w from 0 to 1 000.
• This graph shows a plot of the excess latency versus the corresponding number of NACKs for a group of 1 0,000 receivers for both SPL and EXP timers.
• Minimisation of the objective function ⁇ in which ⁇ (E[M]) is replaced simply by E[M] was performed, using varying values of w from 0 to 1000 (which corresponds to moving along the curves plotted) . Varying the value of the weighting term w corresponds to shifting the relative importance of the latency concern versus the NACK suppression.
• a first step 50 involves generating a lookup table by repeatedly performing the timer function optimization as described with reference to Fig. 2, using various different input conditions and constraints.
• the look-up table may be generated either by the sender itself before multicast commences, or alternatively pre-computed on an external computing system and stored on an electronic storage device accessible by the sender.
• the stored optimised parameters in the lookup table are associated with the relevant input conditions and multicast constraints for which they were generated (for example different parameters may be stored in relation to different group sizes R, or for different types of multicast application).
• the lookup table also stores an indication of which timer function is to be used in a particular situation.
• Multicast is intiated (step 51 ) by the sender selecting from the lookup table the appropriate timer parameters, which will depend on the type of multicast application and anticipated group dynamics.
• the selected parameters (and indication of which timer function to use, if appropriate) are sent by the sender to the receivers in transmitted multicast messages.
• the sender periodically monitors the multicast conditions (step 52).
• a test is applied to determine whether specific multicast conditions have changed by a predefined amount. If not, processing returns to step 52 until the next periodic check is performed during multicasting. However, if the multicast conditions have changed then the optimum timer parameters are re-selected from the lookup table in relation to the updated conditions (step 54).
• the newly selected parameters are then fed back into the multicast for use during the next round (step 55).
• the implementation of the second embodiment is more efficient during run time than the first embodiment since no re-calculation of the timer parameters has to be peformed during multicast, they just have to be re-selected from the lookup table. Accordingly, if it is likely that the multicast conditions will change rapidly, for example if group size R rapidly increases or decreases, then the second embodiment will be preferable.
• a third alternative embodiment comprises a general-purpose computer not itself forming part of a multicast system, running the same processing stages as described and illustrated earlier with reference to Fig. 2.
• This might be used as a modelling tool by, for example, multicast protocol designers, who would enter the required input information, such as the estimates for R and GRTT, using a suitable known input device, such as a keyboard .
• Processing stage 30 is performed by the processor of the computer, with the optimised solution output at step 31 via any suitable output device, such as a display screen, or saved to an electronic storage device.
• the computer of the third embodiment would be used as a tool for either investigating feedback suppression or designing/testing an optimum timer function.
• the output solution may be utilised within a suitable multicast protocol.
• unicast feedback channels having only a single logical direction back to the sender would include satellite networks using a terrestrial unicast feedback channel.
• Feedback supression can occur by either the sender forwarding (by multicast) any NACKs it receives or alternatively multicasting some other indication that a repair request has been received. In the event that a receiver detects either of these before it's backoff time expires then it will suppress it's own feedback.
• the delay before the forwarded NACK or indication is detected by the other receivers will necessarily be longer (transmission time of the order of 2c) compared with when the receivers can directly detect NACKs multicast from other receivers.
• Simple modifications to the transmission times within the expressions for E[X] and E[M] can be made to account for these differences, together with a larger value for timer period T.
• the apparatus that embodies the invention could be a general purpose device having software arranged to provide an embodiment of the invention.
• the device could be a single device or a group of devices and the software could be a single program or a set of programs.
• any or all of the software used to implement the invention can be contained on various transmission and/or storage mediums such as a floppy disc, CD- ROM, or magnetic tape so that the program can be loaded onto one or more general purpose devices or could be downloaded over a network using a suitable transmission medium.

Landscapes

• Engineering & Computer Science (AREA)
• Computer Networks & Wireless Communication (AREA)
• Signal Processing (AREA)
• Data Exchanges In Wide-Area Networks (AREA)
• Mobile Radio Communication Systems (AREA)

Priority Applications (1)

Applications Claiming Priority (4)

Publications (2)

Family

ID=29797294

Family Applications (1)

Country Status (4)

Families Citing this family (37)

Family Cites Families (5)

• 2003
  • 2003-06-12 CA CA2488369A patent/CA2488369C/en not_active Expired - Fee Related
  • 2003-06-12 US US10/517,652 patent/US7526523B2/en not_active Expired - Fee Related
  • 2003-06-12 WO PCT/GB2003/002530 patent/WO2004002048A1/en not_active Ceased
  • 2003-06-12 EP EP03760791.8A patent/EP1516449B1/de not_active Expired - Lifetime

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events